Chris Parsons is the co-founder and chief technology officer of Cherrypick, an AI-powered meal-planning and grocery-shopping company, and the creator of airskills, an open-source system for sharing reusable AI-agent skills. His approach to automation favors small, inspectable workflows, meaningful feedback, and human control over consequential decisions.
From software founder to applied AI
Before Cherrypick, Parsons ran Eden Development, coached software teams, and developed independent games, including Sol Trader. After contracting with Gower Street Analytics on film-analysis and box-office forecasting tools, he became its chief technology officer in late 2017, returning to full-time engineering leadership and data-intensive products. He described that transition in his account of joining Gower Street.
In 2021, Parsons and Tom Foster-Carter launched the grocery-shopping venture originally called Lollipop and later renamed Cherrypick. Its product combines meal planning, supermarket ordering, and personalized recommendations, estimating which ingredients customers already have while allowing them to correct mistaken assumptions. Its production architecture reflects Parsons’s insistence that useful AI requires explanations, feedback, and practical controls.
Parsons also advises and trains executives and engineering teams adopting AI. His writing about startup leadership emphasizes the CEO-CTO relationship as a foundational responsibility for technical founders.
The principles behind his work
Simple autonomous coding loops. Parsons’s Ralph implementation selects one meaningful ticket, implements and verifies the change, records progress, and starts again. At AI Engineer Europe, he demonstrated how durable files and Git history can coordinate successive agent sessions without elaborate dependency graphs or premature multi-agent orchestration.
Fix the actual engineering bottleneck. Applying Agile principles and the theory of constraints, Parsons argues that faster code generation can hurt teams whose real limitations are review capacity, release processes, cross-team dependencies, or unclear requirements. He prefers iterative, just-in-time planning over exhaustive specifications that reproduce Waterfall-style coordination failures.
Portable, versioned agent skills. Through airskills, Parsons packages and synchronizes reusable instructions across coding assistants, with support for sharing, version tracking, conflict detection, and rollback. His writing on skill sharing treats these skills as organizational knowledge that should move between people, machines, and tools.
Human judgment at irreversible boundaries. Parsons uses isolated environments, restricted permissions, separate credentials, automated tests, and review to constrain autonomous agents. His systems can draft emails, prepare presentations, and advance projects, but consequential external actions and security-sensitive production changes remain subject to human approval. He also acknowledges that nondeterministic model behavior makes improvements to agent skills difficult to evaluate rigorously.
His earlier shell-based Pomodoro tracker, p, reflects the same preference for focused tools that solve concrete problems without unnecessary machinery.
